Remove compute capability lazy cell (#2580)

Remove compute capability lock

We are only calling the `get_cuda_capability` function once, so avoiding
the cost of multiple calls is not really necessary yet.

launcher/src/gpu.rs

-use std::sync::LazyLock;
-
-pub static COMPUTE_CAPABILITY: LazyLock<Option<(usize, usize)>> =
-    LazyLock::new(get_cuda_capability);
-
-fn get_cuda_capability() -> Option<(usize, usize)> {
+pub fn get_cuda_capability() -> Option<(usize, usize)> {
     use pyo3::prelude::*;
 
     let py_get_capability = |py: Python| -> PyResult<(isize, isize)> {

launcher/src/main.rs

@@ -66,7 +66,7 @@ fn get_config(
 }
 
 fn resolve_attention(config: &Option<Config>, lora_adapters: &Option<String>) -> (String, String) {
-    let compute_capability = *gpu::COMPUTE_CAPABILITY;
+    let compute_capability = gpu::get_cuda_capability();
     let mut prefix_caching: Option<String> = std::env::var("USE_PREFIX_CACHING").ok();
     let mut attention: Option<String> = std::env::var("ATTENTION").ok();
     if let Some(config) = config {